tahr-secure-app
Pass
Audited by Gen Agent Trust Hub on Jul 25, 2026
Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
- [SAFE]: The skill instructions (SKILL.md and evidence-contract.md) enforce professional security standards, requiring explicit authorization for runtime tests and prohibiting the testing of production or third-party environments.
- [DATA_EXFILTRATION]: The skill implements proactive measures to prevent sensitive data leakage. It requires redacting raw secrets and provides a Python validation script (
scripts/validate_review.py) that uses regular expressions to detect common credential formats (e.g., AWS keys, GitHub tokens, private keys) in the final report artifacts. - [COMMAND_EXECUTION]: The skill includes a local utility script (
scripts/validate_review.py) to ensure review artifacts adhere to the defined schema and safety constraints. The script relies solely on standard Python libraries and performs local validation logic. - [INDIRECT_PROMPT_INJECTION]: The skill identifies a broad ingestion surface as it processes application source code, configuration files, and runtime data. This risk is managed through a structured evidence contract that requires independent proof for all findings and specific remediation targets.
- Ingestion points: Application source code, manifests, lockfiles, and runtime HTTP/API responses identified in the 'Map before judging' section of
SKILL.md. - Boundary markers: The review mode configuration (
deepvsfocused) and explicit authorization scope defined in the report template. - Capability inventory: The skill routes tasks to specialized tools for authentication testing, access control auditing, and input tracing as defined in
references/review-routing.md. - Sanitization: The skill mandates secret redaction and uses
scripts/validate_review.pyto scan for unredacted tokens before final delivery.
Audit Metadata